Sector Integration and the Benefits of Global Diversification

* Dr. Leal thanks the financial support from grants from CNPq and FAPERJ as well as

research grants from the COPPEAD Graduate School of Business.

(Multinational Finance Journal, 2005, vol. 9, no.3/4, pp. 237–269)© Multinational Finance Society, a nonprofit corporation. All rights reserved.DOI: 10.17578/9-3/4-5

1

Sector Integration and the Benefits of GlobalDiversification

Mitchell RatnerRider University, New Jersey, USA

Ricardo P. C. LealCOPPEAD Graduate School of Business, Brazil

One of the main reasons that investment advisors recommend internationalinvestments is that foreign stocks are not highly correlated with U.S. stocks. Asworld economies become increasingly interrelated, it may become more difficultfor investors to achieve effective diversification. This research investigatesinternational stock market correlation, and assesses whether globaldiversification on a sector basis is beneficial to U.S. investors. This analysisincludes 38 developed and emerging stock markets from 1981-2000. In additionto demonstrating a potential loss of diversification benefits, this paper utilizesan optimal global asset allocation model to illustrate the effects of sectordiversification on portfolio performance over time. The results indicate thatalthough the correlation between most foreign sectors and U.S. sectors isincreasing over time, there are still substantial international diversificationbenefits. Further, the inclusion of emerging market sectors may significantlyenhance the return-to-risk performance of international portfolios (JEL: F21,F36, G11, G15).

Keywords: sectors, optimal portfolio, international diversification, co-movement.

I. Introduction

There is a growing concern among both individual and professionalinvestors regarding the benefits of international portfolio diversification.Since the world stock market crash of October 1987, investors areacutely aware that markets are indeed interrelated. Global market

Multinational Finance Journal238

correlations increase during periods of greater economic integration asis apparent during the late 19th and 20th centuries (Goetzmann et al.[2005]). Greater economic integration may be achieved throughincreased trade and cross-border investments. Trade has continued torise dramatically due to the reduction of trade barriers and theproliferation of large trading blocs (e.g., the European Union [EU], andthe North American Free Trade Agreement [NAFTA]). The fall of tradebarriers began with the General Agreement on Tariffs and Trade(GATT), which later produced the World Trade Organization (WTO).These agreements have resulted in increases in economic integration,and the globalization of business enterprise. Economic policycoordination led to a single currency in the EU.

The linkage between international markets increases dramaticallydue to the acceleration of cross-border investments. Factors includingglobal deregulation of the telecommunications, utility, and otherindustries increase competition. Industry consolidations and globalmerger-and-acquisition activity have all helped to strengthen tiesbetween markets worldwide. It is not just the major stock marketindexes (i.e., Dow, Nikkei, FTSE, etc.) that are linked, but alsoindustries and individual firms that are closely tied together. Theglobalization of corporate revenues and expenses, and the growingproportion of intra-industry mergers and acquisitions have greatlyinfluenced the relative importance of sector factors in explainingsecurity returns.

Goetzmann et al. (2005) argue that diversification benefits changethrough time and are driven by either low correlations in the worldmarkets or a large opportunity set. They believe that diversificationbenefits are currently lower than in previous periods during their 150-year sample. However, there have been other periods of lowdiversification benefits, such as in the late 19th century. They suggestthat current diversification benefits are driven mostly by a larger andincreasing opportunity set, because correlations are actually rising.They also attribute an important role to emerging stock markets ascurrent diversification benefits are mostly derived from marginalmarkets. Meric et al. (2001) state that there is no diversification benefitto U.S. investors from investing solely in well-diversified countryindexes in Latin America. They posit that investors would benefit themost from investing in selected industries or securities in thesecountries.

The purpose of this study is to examine the increase in correlation

239Sector Integration

that world markets experience from 1981–2000, and to assess anysubsequent loss of global diversification benefits. The stock indexes of38 developed and emerging countries are subdivided into 10 leadingsector components (e.g., utilities, technology, etc.) to analyze the microlinkage between markets. Building on Goetzmann et al. (2005) andMeric et al. (2001), this study includes sector analysis to offer a broaderarray of investments. The apparent increase in international marketintegration is assessed using correlation and panel data analysis. Paneldata asymmetry analysis is utilized to measure greater integrationbetween markets and sectors during either upturns or downturns in U.S.markets. As correlation is a key factor in determining the benefits ofportfolio diversification, a portfolio optimization model is applied toshow the potential benefit of sector analysis in internationaldiversification. The benefits of international diversification areinvestigated with particular focus on total market investment comparedto sector-based investment in developed and emerging markets.

This paper provides evidence that international investing isbeneficial to U.S. investors, even though this analysis documents thatinternational stock market correlation has increased among the totalstock market indexes of both developed and emerging markets. Micro-market analysis reveals that certain sectors do not experience aconsistent increase in correlation over time, which allows for potentiallygreater diversification benefits. This paper presents evidence comparinginternational investment in total market indexes versus sector-basedinvestment. Utilizing an ex post optimal portfolio model, it is shownthat diversification among international markets using total marketindexes could be superior to investing solely in the U.S. total marketindex. Further, that international sector-based diversification could besuperior to simply holding a diversified portfolio of total marketindexes. The results indicate that fundamental analysis of whichcountries and sectors to include in internationally diversified portfoliosis potentially profitable. Additional findings support the inclusion ofemerging market investments to achieve maximum portfoliodiversification benefits.

II. Background and Literature Review

There is a considerable body of early empirical evidence documentingthe benefits of international portfolio diversification including Levy and


Sarnat (1970) and Solnik (1974). However, recent studies indicate thatcorrelations between the U.S. and most developed equity markets haverisen (Meric and Meric [1998], Longin and Solnik [1995], Erb et al.[1994]), but stabilize after the 1987 crash period (Solnik et al. [1996]).Emerging markets exhibit very low correlations with developed markets(Divecha et al. [1992)], Harvey [1995]), but these correlations areincreasing over time, and appear higher in times of greater internationalvolatility (Erb et al. [1995], Aggarwal and Leal [1997], Bekaert andHarvey [1997], Meric et al. [2001]).

Several studies suggest that the opening of emerging financialmarkets reduces financial market segmentation (Bekaert and Harvey[1997], Bekaert [1995]). Market opening can be achieved through botheconomic and financial reforms. Trade liberalization is among the usualmarket opening economic reforms that have a positive impact on marketvaluations (Henry [2000]). Emerging markets may become moreefficient with trade liberalization as returns show random walkproperties, while financial liberalization does not seem to affectefficiency (Basu and Morey [2000] and Kawakatsu and Morey [1999]).Bekaert and Harvey (2000) find that emerging market correlationincreases with the world market return after financial liberalization.The main attraction of emerging markets to investors is not only thegreater potential returns that can be earned, but that they have low stockmarket correlations with developed markets. As emerging marketsbecome increasingly linked with developed markets, the benefit ofportfolio diversification may diminish.

Most of the prior studies cited focus on the relationship between themajor stock market indexes of each country. Roll (1992) indicates thatindustry concentration is also a significant variable affecting equitymarket correlation. A number of studies investigate the relationshipbetween capital market integration and security returns with someconflicting results. Beckers et al. (1996) examines country and industryfactors, and does not find increasing global integration, except withinthe European Union. Heston and Rouwenhorst (1994) find that sectorsaccounted for less than 4% of the variation in stock return indexes of 12European equity markets. Rouwenhorst (1999) finds that despite theformation of the European Union, individual country effects are stillrelevant.

More recently, Baca et al. (2000) conclude that industrial sectorfactors are increasingly important in explaining national equity returnsin seven major industrial countries (including the U.S.). Serra (2000)


shows that although country effects are the most important factorsexplaining emerging market stock returns, investors should not ignoreindustry effects when they include emerging markets in their portfolios.Miller (2002) believes that both country and sector analysis are nowequally important particularly due to technology. The author says thatglobal sector effects may be confined to a few sectors and that others,such as consumer and industrial stocks, are traded locally. Miller addsthat thinking in terms of country and sector effects is equivalent tothinking locally for some industries and globally for others.

III. Data

The sample consists of U.S. dollar-denominated total monthly indexreturns (including dividends) for 38 countries provided by Datastreamfrom 1981–2000. There are 18 developed countries and 20 emergingcountries. Emerging countries are identified as such by Morgan StanleyCapital International. Using U.S. dollar returns instead of local returnshas the added benefit of accounting for disparate levels of inflation,particularly in some of the emerging countries. The developed samplebegins in 1981, and the emerging sample in 1991 due to the datalimitations of Datastream. Data collected for each country includes thetotal stock market index and 10 sectors within each of the markets. (Insome countries, particularly emerging markets, 10 sectors may not exist.The total stock market index is created by Datastream as a consistentmeasure across all countries in the database.)

Datastream categorizes industries as defined by the Financial TimesActuaries Index into the following sectors: basic industries, cyclicalconsumer goods, cyclical services, financials, general industrials,information technology, nonclyclical consumer goods, nonclyclicalservices, resources, and utilities. The country indexes are weighted bymarket capitalization, contain the largest firms in each market, andrepresent close to 80% of each country’s total market capitalization.There is no overlap between indexes, as foreign listings, includingAmerican Depositary Receipts, are excluded from each index.

All statistical tests are based on the perspective of a U.S. investor.The sample is divided into four 60-month investment horizons to assesschanges over time-period I (January 1981–December 1985), period II(January 1986–December 1990), period III (January 1991–December


TABLE 1. Mean and Standard Deviation for Country Indexes. U.S. Dollar MonthlyReturns (in %).

Series Mean Std dev. Mean Std dev.(1991–2000) (1991–2000) (1981–2000) (1981–2000)

Developed:Australia 0.63 5.27 0.56 6.31Austria 0.28 4.90 0.76 6.83Belgium 0.59 4.29 0.96 5.20Canada 0.87 4.90 0.75 4.85Denmark 0.84 4.59 1.10 5.35Finland 2.19 9.16 1.88 7.85France 0.96 4.97 0.99 6.16Germany 0.68 4.45 0.94 5.32Ireland 1.04 5.28 1.12 6.27Italy 0.53 7.10 0.79 7.41Japan 0.00 6.85 0.72 7.21Netherlands 1.08 4.24 1.16 4.37Norway 0.45 7.13 0.89 7.53Spain 0.63 6.15 1.00 6.62Sweden 1.00 6.20 1.30 6.18Switzerland 1.28 4.62 1.17 4.91U.K. 0.74 4.21 0.92 5.02U.S. 1.23 3.71 1.11 3.85

Emerging:Argentina 1.46 12.10 n/a n/aBrazil 2.49 16.03 n/a n/aChile 0.98 7.36 n/a n/aChina 1.89 11.93 n/a n/aGreece 0.58 9.27 n/a n/aHong Kong 1.27 8.52 n/a n/aIndia 0.32 11.36 n/a n/aIndonesia –1.33 13.46 n/a n/aKorea –0.30 12.40 n/a n/aMalaysia 0.13 11.36 n/a n/aMexico 0.87 10.65 n/a n/aNew Zealand 0.32 6.24 n/a n/aPhilippines 0.53 10.27 n/a n/aPoland –0.80 13.95 n/a n/aPortugal 0.53 6.03 n/a n/aS. Africa 0.21 7.82 n/a n/aSingapore 0.40 6.93 n/a n/aTaiwan 0.16 10.27 n/a n/aThailand –0.58 12.48 n/a n/aTurkey 0.31 18.25 n/a n/a

Note: n/a = not available


1. October 1987 is removed from the analysis as the inordinately high negativecorrelations during that month among stock markets worldwide would bias the findings. Asindicated by Solnik, et. al. (1996), the shock of October 1987 over a multi-decade period ofanalysis is not exceptional. However, in this 5-year analysis, the October 1987 shock ispervasive. For example, the correlation in table 4 for the Total Market Index for 1986–1990is reported as 0.35, which excludes October 1987. If October 1987 is included, the correlationincreases to 0.50.

1995), and period IV (January 1996–December 2000). Data fromOctober 1987 are removed from the analysis.1

Statistics for the total stock market indexes of each country arepresented in table 1. Monthly means and standard deviationsdemonstrate the relative risk-return tradeoff between developed andemerging markets. Although the developed sample spans from1981–2000, the developed sample is also presented during the sametime frame as the emerging sample (1991–2000) for comparisonpurposes. Among developed countries from 1991–2000, Finland(2.19%) has the highest monthly mean and Japan (0.00%) has thelowest. Standard deviation of returns is highest for Finland (9.16%) andlowest for the U.S. (3.71%). In the emerging countries, Brazil (2.49%)has the highest mean, while Indonesia, Korea, Poland, and Thailandexperience negative monthly means. Turkey(18.25%) has the higheststandard deviation and Portugal (6.03%) has the lowest. The standarddeviations indicate that emerging markets have much greater volatilitythan do developed countries during that period of time.

Monthly means and standard deviations are provided for the sectorreturns in table 2. Since there are roughly 380 individual sector series,the data in table 2 report averages of sectors across countries. Thesample is split between developed and emerging countries. Of thedeveloped country sectors from 1991–2000, information technology(1.57%) has the highest mean return and resources (0.32%) has thelowest. The standard deviation is highest for information technology(10.43%) and lowest for utilities (5.91%). Among the emergingcountries, information technology (1.64%) has the highest mean return,while cyclical goods (–0.11%) has the lowest. The standard deviationof information technology (19.35%) is also highest and nonclyclicalgoods (10.58%) has the lowest.

Again, the developed data is presented from 1991–2000 forcomparison purposes with the emerging sample. The fullsample(1981–2000) is also provided for the developed sample.

Some industries are dominated by only a few companies. Indeed,


some country indexes can also be influenced by a major firm (e.g.,Nokia in Finland during the late 1990’s). To examine this issue in moredetail, table 3 contains the number of firms in each sector, by countryas of December 2000. The U.S., U.K. and Japan are the only countrieswith a substantial number of firms in virtually all sectors. Otherdeveloped market sectors contain a range of one firm to several dozenfirms. In the emerging markets, most sectors have fewer than 12 firms.Many of the emerging sectors have only one-to-three firms.International investment, particularly in emerging markets, is subject tothe realities of thinly traded markets, and markets dominated by a fewlarge firms. Portfolio managers should be aware that many foreignsectors may not be adequately diversified. (The optimal portfoliosformed in this analysis contain an 80% base investment in the U.S.,

TABLE 2. Mean and Standard Deviation for Sector Indexes. U.S. Dollar MonthlyReturns (in %).

Series Mean Std dev. Mean Std dev. (1991–2000) (1991–2000) (1981–2000) (1981–2000)

Developed countries:Basic industries 0.33 6.62 0.70 7.39

Cyclical goods 0.59 7.77 0.77 8.77Cyclical services 0.70 6.37 0.95 7.70General Industrials 0.60 6.88 0.76 7.36Information Technology 1.57 10.43 1.26 10.16Noncyclical goods 0.80 6.02 1.13 6.79Noncyclical services 0.91 7.44 1.20 8.01Resources 0.32 7.19 0.63 8.51Financials 0.74 6.82 0.92 7.29Utilities 0.48 5.91 0.75 7.01

Emerging countries:Basic industries 0.06 13.64 n/a n/aCyclical goods –0.11 13.49 n/a n/aCyclical services 0.38 13.08 n/a n/aGeneral Industrials 0.57 15.48 n/a n/aInformation Technology 1.64 19.35 n/a n/aNoncyclical goods 0.47 10.58 n/a n/aNoncyclical services 1.03 12.47 n/a n/aResources 0.40 14.97 n/a n/aFinancials 0.31 12.80 n/a n/aUtilities 0.24 13.00 n/a n/a

.


which avoids the potential of holding a portfolio consisting of only ahandful of equities.)

IV. Methodology and Results

A. Correlations over time

Low correlations between international markets is one of the primereasons for international stock diversification. As the focus of this studyis from the perspective of a U.S. investor, correlations are calculatedbetween individual U.S. sectors and individual foreign country sectorson a country-by-country basis. Since there are close to 380 separateseries (not including the total market series), sector correlations areaveraged across countries. For example, U.S. basic industries arecorrelated against the average of the remaining industries (i.e., Australiabasic, Austria basic, Belgium basic, ...). The average between-countrysector correlations for four 60-month investment periods are given intable 4.

Several conclusions can be drawn from the results. The averagecorrelation of the U.S. total market with other developed markets issteadily increasing from 0.31 in 1981–1985, to 0.59 in 1996–2000. Onthe surface, this dramatic increase in correlation may indicate a potentialloss in diversification benefits. The sector correlations are not consistentover time. The information technology sector has stable correlationsuntil the last period, while most other sectors show some variationbetween periods. However, the fourth period correlations are typicallytwo or three times higher than those in the first period in eight of the tensectors. The two notable exceptions are the resource sector with fairlystable correlations, and the utilities sector with very low correlations.

The trend in correlations between the U.S. and emerging markets aresimilar to the developed markets from 1991–2000. The correlationbetween the U.S. total market and the average emerging total marketindex increases from 0.20 (1991–1995) to 0.43 (1996–2000), which alsoindicates a potential overall loss of international diversification benefitsrelative to correlation. The sector correlations are generally highest inthe fourth period, although certain industries demonstrate consistentcorrelations between the two periods (i.e., cyclical and nonclyclicalgoods, utilities).

In sum, the rising correlations indicate a potential loss in inter-


TA

BL

E 3

. Num

ber

of F

irm

s in

Eac

h Se

ctor

, by

Cou

ntry

, as

of D

ecem

ber

2000

.

Ser

ies

Res

ourc

eB

asic

Gen

eral

Ind

.C

yc. G

oods

Non

cycl

.Goo

dsC

yc. S

erv.

Non

ycl.

Ser

v.U

tili

ties

Info

. Tec

hF

inan

cial

s

Aus

tral

ia17

177

219

364

52

51A

ustr

ia1

108

34

21

21

18B

elgi

umn/

a15

102

129

52

629

Can

ada

5529

147

2140

1111

1052

Den

mar

kn/

a7

72

139

11

19

Fin

land

111

112

68

41

24

Fra

nce

717

2520

4063

131

2736

Ger

man

y1

2743

2724

329

1422

49Ir

elan

d4

81

212

91

n/a

58

Ital

y4

1923

197

187

95

49Ja

pan

1215

318

590

118

166

3217

6216

4N

ethe

rlan

ds7

1214

1112

267

n/a

1130

Nor

way

83

72

212

22

39

Spa

in2

2410

712

215

82

29S

wed

enn/

a11

182

76

51

218

Sw

itze

rlan

dn/

a18

264

2314

36

154

U.K

.20

4741

846

173

1613

2216

1U

.S.

5058

7149

152

178

2864

105

243

(Con

tinu

ed)


TA

BL

E 3

. (C

onti

nued

)

Ser

ies

Res

ourc

eB

asic

Gen

eral

Ind

.C

yc. G

oods

Non

Cyc

.Goo

dsC

yc. S

erv.

Non

Cyc

. Ser

v.U

tili

ties

Info

. Tec

hF

inan

cial

s

Arg

enti

na1

13

27

34

7n/

a11

Bra

zil

720

84

71

2416

n/a

13C

hile

19

6n/

a8

62

9n/

a9

Chi

na5

104

82

11n/

a6

22

Gre

ece

311

n/a

15

93

23

13H

ong

Kon

g2

320

119

304

57

39In

dia

518

166

215

24

1013

Indo

nesi

a2

61

69

73

n/a

n/a

16K

orea

215

1910

1012

52

322

Mal

aysi

a3

116

516

134

83

21M

exic

o1

1610

119

239

n/a

110

N. Z

eala

nd1

61

26

162

4n/

a16

Phi

lipp

ines

11

3n/

a7

75

4n/

a22

Pol

and

17

23

94

21

419

Por

tuga

ln/

a10

32

314

71

19

S. A

fric

a16

73

110

104

n/a

n/a

19S

inga

pore

25

202

1324

21

526

Tai

wan

n/a

55

41

43

n/a

3216

Tha

ilan

d2

102

n/a

28

32

219

Tur

key

37

711

25

24

n/a

9


2. We also examine the distribution of the correlation coefficients found in table 4, asthe reported correlations are a function of the number and selection of the countries includedin the study. For example, the reported total market index correlation is 0.59 for the1996–2000 sub period. The range is from 0.45 to 0.79 when examining individual totalmarket index correlations by country (results not reported here). The mean and median areboth 0.59, indicating a symmetrical distribution of the data. The standard deviation of 0.09

national diversification benefits on a total market basis, but sectorinvesting still may offer effective benefits due to consistent or lowcorrelations. Further, while emerging market correlations are increasingover time, the level of correlation with the U.S. market remains lowerfor emerging markets compared to developed markets.2

TABLE 4. Average Correlation of U.S. Market/Sectors with Developed andEmerging Markets/Sectors.

1981– 1986– 1991– 1996–1985 1990 1995 2000

Developed Countries:Basic industries 0.28 0.35 0.35 0.51Cyclical goods 0.12 0.29 0.26 0.32Cyclical services 0.18 0.32 0.15 0.37General Industrials 0.20 0.35 0.31 0.52Information Technology 0.25 0.25 0.24 0.50Noncyclical goods 0.25 0.30 0.25 0.36Noncyclical services 0.04 0.25 0.14 0.42Resources 0.48 0.31 0.42 0.43Financials 0.23 0.20 0.24 0.48Utilities 0.09 0.12 0.20 0.02

Total market index(developed countries) 0.31 0.35 0.41 0.59Emerging Countries:

Basic industries n/a n/a 0.17 0.32Cyclical goods n/a n/a 0.13 0.16Cyclical services n/a n/a 0.07 0.28General Industrials n/a n/a 0.20 0.38Information Technology n/a n/a 0.07 0.39Noncyclical goods n/a n/a 0.16 0.20Noncyclical services n/a n/a 0.05 0.30Resources n/a n/a 0.16 0.27Financials n/a n/a 0.03 0.29Utilities n/a n/a 0.04 0.05

Total market index(emerging countries) n/a n/a 0.20 0.43

Note: For example, Basic industries (0.28) represents the average of the U.S. Basicindustries with the Basic industries of each individual country during the first period.


indicates a relatively tight dispersion. Thus, among developed markets, no country appearsto bias the overall correlation structure. For the emerging total market sample, the mean(0.43) and median (0.45) are also fairly close together indicating a generally symmetricaldistribution. The standard deviation is 0.13, indicating a somewhat larger variation than thatof the developed sample, which is not unexpected. The details of the distributions of the 10sectors for both the developed and emerging sample are not reported here, but indicatevarying degrees of dispersion of correlation coefficients. To avoid unintentional data mining,all available sectors are included in the analysis.

B. Panel Data Analysis

To study the effects of time variability and to increase the efficiency ofthe parameter estimates, cross-sectional and time series data are pooledto form a panel data set. There are several advantages to using paneldata. First, panel data allows the examination of the relationshipbetween the U.S. sectors and all foreign sectors over time in amulti-country framework. Second, panel data provides additional datapoints that increase degrees of freedom. Third, utilizing both cross-section and time series data may reduce problems that can occur due toomitted-variables.

Panel data does introduce statistical difficulties in modelspecification as the error term may contain time series disturbances,cross-section disturbances, or both. The Durbin-Watson statistic foreach regression is examined to test for time series disturbances (serialcorrelation). In addition, a random-effects model is utilized that allowsfor the error term to be correlated over time and across countries, whichaccounts for cross-sectional disturbances.

The basic framework for the panel data model is the generalizedregression model:

(1)it i it itγ β χ ε= +, ,i t i t i tu v wε = + +

assuming that:ui -Ν †0, σ2

u is the cross-section error componentvt -Ν †0, σ2

v is the time series error componentwi,t -Ν †0, σ2

w is the combined error component

Pooling is achieved by stacking n-time series so that:


TA

BL

E 5

. G

ener

aliz

ed L

east

Squ

ares

(G

LS)

Est

imat

es f

rom

Dev

elop

ed F

orei

gn M

arke

ts/S

ecto

rs P

anel

Reg

ress

ed o

n U

.S.

Mar

kets

/Sec

tors

usi

ng a

Ran

dom

–Eff

ects

Mod

el.

Bet

a C

oeff

icie

nts

and

Adj

uste

d R

2 Rep

orte

d. (

t–st

atis

tics

are

in

Par

enth

eses

). C

how

Tes

t fo

r St

ruct

ural

Sta

bilit

y be

twee

n 19

91–1

995

and

1996

–200

0.

1981

–198

519

86–1

990

1991

–199

519

96–2

000

Cho

w F

–sta

tS

ecto

r:be

taad

j.R2

Bet

aad

j.R2

beta

adj.R

2be

taad

j.R2

Bas

ic in

dust

ries

0.39

***

0.06

0.45

***

0.10

0.63

***

0.10

0.54

***

0.25

3.90

***

(8.1

0)(1

0.36

)(1

1.16

)(1

8.67

)C

ycli

cal g

oods

0.21

***

0.01

0.50

***

0.09

0.43

***

0.07

0.38

***

0.09

3.87

***

(2.6

5)(8

.52)

(6.6

0)(9

.18)

Cyc

lica

l ser

vice

s0.

27**

*0.

020.

40**

*0.

080.

26**

*0.

020.

47**

*0.

136.

85**

*(4

.04)

(9.2

2)(4

.58)

(12.

07)

Gen

eral

Ind

ustr

ials

0.29

***

0.04

0.50

***

0.11

0.63

***

0.08

0.74

***

0.25

5.78

***

(5.9

2)(1

0.99

)(9

.62)

(18.

56)

Info

. Tec

hnol

ogy

0.38

***

0.05

0.30

***

0.07

0.45

***

0.06

0.64

***

0.25

6.46

***

(3.7

5)(5

.67)

(6.1

5)(1

5.20

)

(Con

tinu

ed)


TA

BL

E 5

. (C

onti

nued

)

1981

–198

519

86–1

990

1991

–199

519

96–2

000

Cho

w F

–sta

tS

ecto

r:be

taad

j. R

2B

eta

adj.

R2

beta

adj.

R2

beta

adj.

R2

Non

cycl

ical

goo

ds0.

45**

*0.

040.

38**

*0.

070.

36**

*0.

040.

49**

*0.

103.

02**

(6.3

0)(7

.88)

(6.8

1)(1

0.86

)N

oncy

clic

al s

ervi

ces

0.07

0.00

0.34

***

0.03

0.28

***

0.00

0.70

***

0.16

6.86

***

(0.7

4)(5

.25)

(3.3

9)(1

2.04

)R

esou

rces

0.76

***

0.17

0.41

***

0.04

0.73

***

0.12

0.56

***

0.20

4.43

***

(10.

23)

(6.1

8)(1

1.11

)(1

3.13

)F

inan

cial

s0.

29**

*0.

030.

22**

*0.

020.

40**

*0.

030.

52**

*0.

226.

39**

*(5

.86)

(5.3

4)(6

.51)

(16.

19)

Uti

liti

es0.

110.

000.

24**

*0.

000.

42**

*0.

030.

020.

009.

08**

*(1

.00)

(3.0

0)(5

.40)

(0.3

4)

Tot

al M

arke

t Ind

ex0.

54**

*0.

050.

68**

*0.

090.

74**

*0.

120.

73**

*0.

335.

01**

*(d

evel

oped

cou

ntri

es)

(10.

28)

(16.

37)

(13.

93)

(22.

64)

Not

e: *

**, *

*, a

nd *

indi

cate

sig

nifi

canc

e at

the

1%, 5

%, a

nd 1

0% le

vels

, res

pect

ivel

y


1 1 1 1

2 2 2 2

n n n n

y x

y x

y x

β εβ ε

β ε

⎡ ⎤ ⎡ ⎤ ⎡ ⎤ ⎡ ⎤⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥= +⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎢ ⎥ ⎢ ⎥ ⎢ ⎥ ⎢ ⎥⎣ ⎦ ⎣ ⎦ ⎣ ⎦ ⎣ ⎦

The panel data model in the study is empirically estimated as ageneralized least squares (GLS) regression:

(2), , , , ,i t i t i t i t i tFOR USα β ε= + +

where FOR represents the foreign sector returns (in U.S. currency), andUS represents the U.S. sector returns for individual sector i over timeperiod t.

This procedure requires that the observations are weighted inverselyto their variances. As the error component variances are unknown, athree-stage process is performed. The first stage pools together theentire sample based on ordinary least squares, where the residuals aredecomposed into their random and individual components. Stage twocomputes the GLS covariance matrix to determine the precision of theoverall estimates. In the final stage a matrix-weighted average of theindividual estimates are used to calculate the grand coefficient matrix.(A detailed explanation is provided in Greene [1990]).

The regressions are performed on a sector-by-sector basis, andindicate the relationship between the U.S. sector and the cross-sectionalcomparable foreign sector over four 60-month investment periods. Thesample is split between developed countries and emerging countries fortwo reasons. First, to maintain the continuity of the developed samplethat begins 10 years earlier than the emerging sample. Second, to focuson the unique relationship between the U.S. and emerging markets.Beta coefficients, significance levels, and adjusted R2 are reported.

Table 5 contains the results of the foreign sector returns panel-regressed on the U.S. sector returns for the developed countries only.Several observations are apparent from the results. First, the relationshipbetween each U.S. sector and their corresponding foreign sectors are notsimilar within specific time periods. For example, the betas betweensectors vary from a statistically insignificant 0.07 (noncyclical services)to a significant 0.76 (resource) during the 1981–1985 period. Therelatively larger and more significant the beta coefficient, the closer is


the relationship between the individual U.S. sector and thecorresponding foreign sectors. Second, sector betas are not necessarilyconsistent over time. That is, some sectors experience fairly stable betasacross time periods such as noncyclical goods, while other sectors havemuch wider variation (information technology and cyclical services).Third, there is somewhat of an upward trend in the level of the betacoefficients over time, which is especially evident when comparing theperiod 1981–1985 with the 1996–2000 period.

The fourth observation is that the adjusted R2 are noticeably largerin the last period (1996–2000) than in the prior periods for all sectorsexcept the utilities sector. This demonstrates the rising percentage invariation of foreign sector returns explained by U.S. sector returns. Insome cases, the percentage difference is small, such as the R2 in cyclicalgoods between 1991–1995 (0.07) and 1996–2000 (0.09). For mostsectors the difference in adjusted R2 between the third and fourthperiods is substantially larger as in non-cyclical services (0.00[1991–1995] increases to 0.16 [1996–2000]). A Chow test is performedto detect a significant structural change in the model between theperiods 1991–1995 and 1996–2000. The F-statistics reported in table 5reject the null hypothesis that the models are statistically the samebetween periods. All of the F-statistics are significant at the 1% level,with the exception of noncyclical goods significant at the 5% level.

The last row of table 5 contains the results of the foreign totalmarket indexes panel regressed on the U.S. total market index. The totalstock market index is a rough proxy for a well-diversified equityinvestor. The index betas rise from 0.54 (1981–1985) to 0.73(1996–2000). More telling is the rise in adjusted R2 from 0.05 in thefirst period to 0.33 in the fourth period. That is, there is a significant risein the explanatory power of the U.S. total market of foreign totalmarkets during the sample period. Although the betas are similar in thethird and fourth periods, a Chow test indicates a significant structuralchange in the 1996–2000 period.

The total market index betas and R2 are larger in magnitude thanthose of the individual sectors. This may indicate potentially lowerdiversification benefits of international total market investmentcompared with individual sector investment. Sector selection must becarefully made, as some sectors have closer ties to the U.S. in certainperiods. For example, the resource sector in the first period has an R2 of0.17, compared with the R2 of the total market index of 0.05. However,even an R2 of 0.33 for the total market index in the most recent period


is still low enough to potentially offer international diversificationbenefits.

The relationship between the U.S. sectors and emerging marketsectors is examined in table 6. The emerging market data is limited totwo 60- month periods from 1991–2000. Compared with the developedsample during 1991–1995, the emerging sample beta coefficients arelower in magnitude and significance levels. Every sector beta in thedeveloped sample is significant at the 1% level, while only four out often emerging market sector betas are significant during the same period.However, during 1996–2000, all of the emerging beta coefficients are

TABLE 6. Generalized Least Squares (GLS) Estimates from Emerging ForeignMarkets/Sectors Panel Regressed on U.S. Markets/Sectors using aRandom–Effects Model. Beta Coefficients and Adjusted R2 Reported.(t–statistics are in Parentheses). Chow Test for Structural Stabilitybetween 1991–1995 and 1996–2000.

1991–1995 1996–2000 Chow F–statSector: beta adj. R2 beta adj. R2

Basic industries 0.41*** 0.00 0.66*** 0.10 9.90***(3.54) (11.47)

Cyclical goods 0.25*** 0.00 0.31*** 0.01 3.05**(2.55) (4.45)

Cyclical services 0.06 0.00 0.65*** 0.06 11.25***(0.49) (6.65)

General Industrials 0.44*** 0.00 0.98*** 0.07 7.68***(3.38) (10.17)

Info. Technology 0.30 0.03 0.78*** 0.14 9.18***(1.53) (9.41)

Noncyclical goods 0.38*** 0.01 0.44*** 0.03 9.49***(4.11) (6.61)

Noncyclical services 0.09 0.00 0.73*** 0.08 12.63***(0.66) (9.39)

Resources 0.06 0.00 0.61*** 0.06 8.67***(0.36) (8.18)

Financials 0.16 0.00 0.53*** 0.07 15.98***(1.57) (9.41)

Utilities 0.01 0.00 0.16 0.00 6.50***(0.02) (1.53)

Total market index 0.70*** 0.03 0.98*** 0.16 15.88*** (emerging countries) (5.33) (15.31)

Note: ***, **, and * indicate significance at the 1%, 5%, and 10% levels, respectively


significant at the 1% level except for the utilities sector. Likewise, theadjusted R2 are all close to zero during 1991–1995, but rise to moremeasurable levels during 1996–2000 in most cases. The Chow testindicates a significant structural change in the model between the twotime periods.

The emerging total stock market indexes panel regressed on the U.S.total market index indicate a rise in the explanatory power of the U.S.total market over time. The R2 increases from 0.03 (1991–1995) to 0.16(1996–2000). A Chow test confirms a significant structural change.Similar to the developed markets, the emerging R2 in the last period(1996–2000) is higher than the individual sectors. Thus, the potentialbenefits of emerging market investments are likely higher on a sectorbasis rather than a country basis, which is consistent with thepredictions by Meric et al. (2001) and Serra (2000). One exception isthe information technology sector, which has similar R2 statisticscompared with the total market index over time.

In sum, the panel regressions measure the cross-sectional and timeseries relationship of the U.S. markets’ explanatory power of foreignmarkets. From the perspective of a U.S. investor, the more that U.S.sectors explain movements of foreign sectors, the less value the foreignsectors provide in diversification benefits. While there is somevariability in the developed sample beta coefficients during the fourinvestment periods, the adjusted R2 and t-statistics are generally highestin the most recent investment period. The low beta coefficients and R2

during 1991–1995 in the emerging sample illustrate a potentially largeportfolio diversification benefit. The rising magnitudes, significancelevels, and R2 in the emerging sample indicate that the diversificationbenefits of emerging market investment may diminish over time.Although the U.S. sectors appear closer to the foreign sectors in manycases in the most recent period, the betas and R2 are still low enough topotentially provide international diversification benefits.

C. Asymmetry Analysis

One shortcoming of the prior tests is that the estimated coefficients donot depend on the sign of the coefficients, i.e., changes in U.S. stockreturns are assumed to have symmetrical effects on foreign stockreturns. Erb et al. (1995) demonstrate that correlation and volatilitybetween major stock indexes is higher in U.S. down markets. In orderto detect asymmetrical relationships among international sectors, define


TA

BL

E 7

. A

sym

met

ry A

naly

sis.

Gen

eral

ized

Lea

st S

quar

es (

GL

S) E

stim

ates

fro

m D

evel

oped

For

eign

Mar

kets

/Sec

tors

Pan

elR

egre

ssed

on

Pos

itiv

e an

d N

egat

ive

Mov

emen

ts o

f U.S

. Mar

kets

/Sec

tors

usi

ng a

Ran

dom

–Eff

ects

Mod

el.

PO

S &

NE

GC

oeff

icie

nts,

and

F–S

tati

stic

s fo

r E

qual

ity

Tes

ts (

PO

S=N

EG

) re

port

ed.

1981

–198

519

86–1

990

Sec

tor:

PO

SN

EG

PO

S'N

EG

PO

SN

EG

PO

S'N

EG

Bas

ic in

dust

ries

0.43

***

0.33

***

0.33

0.10

0.95

***

31.4

2***

Cyc

lica

l goo

ds0.

130.

37*

0.69

0.04

0.99

***

20.6

3***

Cyc

lica

l ser

vice

s0.

49**

*0.

216.

35**

*0.

17**

*0.

66**

*11

.17*

**G

ener

al I

nd.

0.40

***

0.15

1.89

0.14

0.88

***

23.4

5***

Info

. Tec

hnol

ogy

0.38

**0.

370.

000.

17*

0.44

***

2.40

Non

cycl

. goo

ds0.

61**

*0.

162.

72*

0.35

***

0.44

***

0.36

Non

cycl

. .se

rvic

es0.

01*

0.17

0.22

0.10

0.56

***

4.90

**R

esou

rces

0.84

***

0.68

***

0.40

0.28

***

0.63

***

3.16

*F

inan

cial

s0.

37**

*0.

121.

890.

050.

42**

*6.

54**

*U

tili

ties

0.96

***

–1.0

5***

27.4

5***

0.46

***

–0.0

23.

33*

Tot

al m

arke

t ind

ex(d

evel

oped

cou

ntri

es)

0.42

***

0.65

***

1.36

0.34

***

0.66

***

4.29

**

(Con

tinu

ed)


TA

BL

E 7

. (C

onti

nued

)

1991

–199

519

96–2

000

Sec

tor:

PO

SN

EG

PO

S=N

EG

PO

SN

EG

PO

S=N

EG

Bas

ic in

dust

ries

0.62

***

0.65

***

0.02

0.39

***

0.74

***

15.8

8***

Cyc

lica

l goo

ds0.

21*

0.79

***

6.03

***

0.36

***

0.39

***

0.81

Cyc

lica

l ser

vice

s0.

25**

*0.

29**

0.04

0.45

***

0.49

***

0.07

Gen

. Ind

ustr

ials

0.64

***

0.61

***

0.02

0.56

***

1.08

***

13.8

9***

Info

. Tec

hnol

ogy

0.45

***

0.43

**0.

000.

52**

*0.

77**

*2.

91*

Non

cycl

. goo

ds0.

44**

*0.

162.

070.

67**

*0.

26**

*7.

18**

*N

oncy

cl. s

ervi

ces

0.18

0.41

**0.

660.

66**

*0.

74**

*0.

14R

esou

rces

0.74

***

0.72

***

0.01

0.57

***

0.54

***

0.04

Fin

anci

als

0.24

**0.

66**

*3.

76**

0.43

***

0.62

***

3.75

**U

tili

ties

0.50

***

0.33

**0.

47–0

.01

0.07

0.21

Tot

al m

arke

t ind

ex(d

evel

oped

cou

ntri

es)

0.76

***

0.98

***

0.86

0.57

***

0.94

***

9.99

***

Not

e: *

**, *

*, a

nd *

indi

cate

sig

nifi

canc

e at

the

1%, 5

%, a

nd 1

0% le

vels

, res

pect

ivel

y.


two series (POS and NEG) that contain only positive and negativechanges in U.S. stock sector returns (USSTK), respectively:

( )( )

.... 0

0................. 0

POS if USSTKPOS

if USSTK

⎧ > ⎫⎪ ⎪= ⎨ ⎬≤⎪ ⎪⎩ ⎭

( )( )

.... 0

0................. 0

NEG if USSTKNEG

if USSTK

⎧ < ⎫⎪ ⎪= ⎨ ⎬≥⎪ ⎪⎩ ⎭

Asymmetry tests are then conducted using a GLS panel regression onthe following model:

(3), , , , , , ,i t i t i t i t i t i t i tFOR POS NEGα β γ ε= + + +

POS and NEG coefficients, equality tests, and significance levels for thedeveloped country sample are provided in table 7. The equality testsprovide an F-statistic which tests the null hypothesis that thecoefficients are symmetrical, Ho: POS'NEG. The coefficients varygreatly across sectors and over time, but are generally within the rangeof 0.00 to 1.00. The larger the relative magnitude and significance levelsof the coefficients, the closer the relationship between the U.S. andforeign sectors. As this relationship becomes closer, the benefits ofinternational diversification may diminish. Out of the 40 equationsestimated (10 industries × 4 periods), 18 equations demonstratestatistically significant asymmetry.

The asymmetrical effects between the U.S. total market index andthe foreign total market indexes within the developed sample areprovided in the last row of table 7. All of the positive and negativecoefficients are significant during each period, but the relativemagnitude of the negative coefficients is consistently higher. Theequality tests only indicate a significant difference between the positiveand negative coefficients during 1986–1990 and 1996–2000. In sum, itappears that the correlation between the U.S. and foreign markets isgenerally higher during downturns in the U.S. market. However, manysectors (noncyclical services and resources) provide little or no evidenceof asymmetry. Depending on the time period, it may be possible to


minimize higher overall correlations between international stock returnsdue to downturns in the U.S. market by investing on a sector basis. Itshould be noted that severe downturns (e.g. 1987 crash) are not tested,which are probably unavoidable in all markets and sectors.

Asymmetry analysis for the emerging market sectors is presented intable 8. While several of the POS and NEG coefficients are significantduring 1991–1995, the equality tests indicate that asymmetry exists inonly two sectors. The noncyclical services and financials sectorsdemonstrate a significant response to downturns in the correspondingU.S. sectors. The correlation between U.S. sectors and most emergingsectors does not appear to increase in either up or down movements inU.S. sectors during this time period. The most recent time period(1996–2000) indicates a substantial increase in the magnitude andsignificance levels of most POS and NEG coefficients. Of the four casesof significant equality tests (POS'NEG), the correlation between U.S.and emerging sectors is always higher during downturns in the U.S.sectors than upturns.

The emerging total market indexes are panel regressed on the U.S.total market index to test for asymmetry. The findings in the last row oftable 8 show that emerging total market indexes are significantly relatedto the U.S. total market during both downturns and upturns in the U.S.market. Equality tests indicate that the correlation between emergingmarkets and the U.S. is higher during downturns in the U.S. marketrelative to that during upturns. The emerging market results areconsistent with the developed market results; correlations between U.S.and foreign sectors are generally higher in the most recent period(1996–2000) during downturns in the U.S. market. Compared to thedeveloped markets, the emerging sample contains more sectors that donot have an asymmetrical effect. That is, there are potentially greaterinternational diversification benefits among emerging sectors that areless correlated with U.S. sectors during downturns in the U.S.However, based on the limited emerging sample period (1991–2000),the correlations between emerging and U.S. sectors appear to beincreasing over time.

D. Optimal Sector Allocation

It is possible that arbitrarily selecting foreign sectors or country indexesmay offer some diversification benefits. Even a random selection ofstocks will reduce portfolio risk. Of course, professional investors do


TA

BL

E 8

. A

sym

met

ry A

naly

sis.

Gen

eral

ized

Lea

st S

quar

es (

GL

S) E

stim

ates

fro

m E

mer

ging

For

eign

Mar

kets

/Sec

tors

Pan

elR

egre

ssed

on

Pos

itiv

e an

d N

egat

ive

Mov

emen

ts o

f U.S

. Mar

kets

/Sec

tors

Usi

ng a

Ran

dom

–Eff

ects

Mod

el.

PO

S &

NE

GC

oeff

icie

nts,

and

F–S

tati

stic

s fo

r E

qual

ity

Tes

ts (

PO

S=N

EG

) re

port

ed.

1991

–199

519

96–2

000

Sec

tor:

PO

SN

EG

PO

S=N

EG

PO

SN

EG

PO

S=N

EG

Bas

ic in

dust

ries

0.31

0.58

**0.

430.

53**

*0.

85**

*3.

10*

Cyc

lica

l goo

ds0.

26–0

.05

0.40

0.18

0.23

0.01

Cyc

lica

l ser

vice

s–0

.03

0.22

0.28

0.66

***

0.64

***

0.00

Gen

. Ind

ustr

ials

0.32

0.70

**0.

650.

68**

*1.

51**

*5.

92**

*In

fo. T

echn

olog

y0.

67**

–0.4

22.

080.

57**

*1.

03**

*2.

56N

oncy

cl. g

oods

0.53

***

0.07

1.67

0.36

***

0.54

***

0.63

Non

cycl

. ser

vice

s–0

.34

0.61

**4.

34**

0.45

***

1.03

***

4.60

**R

esou

rces

0.19

–0.1

10.

280.

75**

*0.

40**

1.80

Fin

anci

als

–0.0

70.

57**

2.93

*0.

25**

*0.

86**

*10

.10*

**U

tili

ties

0.33

–0.4

11.

300.

180.

120.

03T

otal

mar

ket i

ndex

(em

ergi

ng c

ount

ries

)0.

42**

*1.

26**

*2.

83*

0.55

***

1.50

***

17.1

9***

Not

e: *

**, *

*, a

nd *

indi

cate

sig

nifi

canc

e at

the

1%, 5

%, a

nd 1

0% le

vels

, res

pect

ivel

y.


not select stocks at random. To demonstrate the potential benefit offundamental analysis for international sector allocation, optimalefficient portfolios are formed over four 60-month investment periodsfrom 1981–2000. As this procedure is performed on an ex post basis,the selected assets are not recommendations for future investment. Thepurpose of this procedure is to illustrate the benefits of internationalsector investments relative to U.S. sector investments and country indexinvestments over time.

Markowitz mean-variance (MV) optimization is used to obtain theoptimal portfolios. The model for portfolio optimization is based on thefollowing:

(4)( )p

p

E rMAX

σΘ =

subject to: ( ) ( )1

N

p i ii

E r x E r=

=∑

1

1N

ii

x=

=∑and

0, 1,ix i N≥ = …

where E(rp) represents the expected return of the portfolio, σp is theportfolio standard deviation, xT is the transpose of a vector of riskyassets weights, and S is the sample variance-covariance matrix. Theportfolio is MV efficient for a given level of portfolio expected return.The model does not allow for short sales or risk free investments. As aresult, the efficient portfolio weights are further constrained to sum to1.0 and to have nonnegative values. The efficient frontier is computedusing 500 efficient portfolios. The investments that maximize theportfolio return-to-risk ratio (MAX Θ) are reported.

The results for six variations of optimized portfolios are presentedin table 9. Four of the variations are constrained to invest 80% in theU.S. market to mimic the allocation of an average U.S. pension fund.Restricting the portfolio to invest 80% in the U.S. ensures that sufficientdiversification is maintained (as noted in data section, numerous foreign


TA

BL

E 9

. Su

mm

ary

Stat

isti

cs f

or M

arko

wit

z M

ean–

Var

ianc

e E

ffic

ient

Por

tfol

io O

ptim

izat

ion.

Com

pari

son

of M

arke

t an

dSe

ctor

–Bas

ed I

nves

tmen

t Str

ateg

ies

in D

evel

oped

and

Em

ergi

ng M

arke

ts. E

mer

ging

Mar

ket d

ata

is u

nava

ilabl

e pr

ior

to 1

991

(n/a

).

POR

TFO

LIO

CO

MPO

SIT

ION

Tot

al M

arke

tT

otal

Mar

ket

Tot

al M

arke

tIn

dexe

sS

ecto

rsS

ecto

rs (

U.S

.,P

ortf

olio

Att

ribu

tes

Ind

ex I

ndex

es(D

evel

oped

&S

ecto

rs(U

.S. &

(Dev

elop

ed &

(all

in %

)(U

.S. o

nly)

(Dev

elop

ed)

Em

ergi

ng)

(U.S

. onl

y)D

evel

oped

)E

mer

ging

)

1996

–200

0M

ean

1.30

1.63

1.50

1.37

1.75

1.91

Sta

ndar

d de

viat

ion

4.63

4.97

3.97

3.43

3.24

2.70

Max

imum

ret

urn/

risk

rat

io28

.08

32.7

037

.83

39.8

154

.09

70.8

5%

inve

sted

in U

.S.

100.

0080

.00

80.0

010

0.00

80.0

080

.00

% in

Dev

elop

ed20

.00

1.03

20.0

01.

20%

in E

mer

ging

18.9

719

91–1

995

Mea

n1.

161.

261.

401.

501.

851.

68S

tand

ard

devi

atio

n2.

512.

432.

502.

882.

691.

88M

axim

um r

etur

n/ri

sk r

atio

46.2

252

.08

56.0

552

.19

68.8

189

.16

% in

vest

ed in

U.S

.10

0.00

80.0

080

.00

100.

0080

.00

80.0

0%

in d

evel

oped

20.0

06.

5720

.00

1.46

% in

Em

ergi

ng13

.43

18.5

4

(Con

tinu

ed)


TA

BL

E 9

. (C

onti

nued

)

POR

TFO

LIO

CO

MPO

SIT

ION

Tot

al M

arke

tT

otal

Mar

ket

Tot

al M

arke

tIn

dexe

sS

ecto

rsS

ecto

rs (

U.S

.,P

ortf

olio

Att

ribu

tes

Ind

ex I

ndex

es(D

evel

oped

&S

ecto

rs(U

.S. &

(Dev

elop

ed &

(all

in %

)(U

.S. o

nly)

(Dev

elop

ed)

Em

ergi

ng)

(U.S

. onl

y)D

evel

oped

)E

mer

ging

)

1986

–199

0M

ean

0.72

0.96

n/a

1.57

1.81

n/a

Sta

ndar

d de

viat

ion

5.43

4.84

n/a

5.67

5.00

n/a

Max

imum

ret

urn/

risk

rat

io13

.26

19.7

8n/

a27

.68

36.1

0n/

a%

inve

sted

in U

.S.

100.

0080

.00

100.

0080

.00

% in

Dev

elop

ed20

.00

20.0

019

81–1

985

Mea

n0.

830.

96n/

a1.

171.

67n/

aS

tand

ard

devi

atio

n3.

603.

33n/

a2.

882.

53n/

aM

axim

um r

etur

n/ri

sk r

atio

23.0

528

.84

n/a

40.6

366

.17

n/a

% in

vest

ed in

U.S

.10

0.00

80.0

010

0.00

80.0

0%

in D

evel

oped

20.0

020

.00


sectors may contain a limited number of equities). The remaining twovariations invest 100% in the U.S.,and are provided for comparisonpurposes only.

Referring to the most recent period (1996–2000), it is clear that thereturn/risk ratios are increasing across variations of the model. Forcomparison purposes, the first column on the left-hand side provides thereturn/risk profile for a 100% investment in the U.S. total market index.The mean (1.30%) and standard deviation (4.63%) produce a return/riskratio of 28.08%. The second variation allows for 80% investment in theU.S. total market index, and 20% in other developed total marketindexes.(There are 17 remaining developed market indexes that may beincluded in the 20% asset allocation). The return/risk ratio is 32.70%,which is an improvement in performance from the 100% U.S. totalmarket index portfolio.

The third variation constrains 80% investment in the U.S. totalmarket index, but allows 20% in foreign total market indexes selectedfrom 17 developed and 20 emerging market indexes. The return/riskratio increases to 37.83%. The fourth variation is an optimized portfolioallocated among 10 U.S. sectors only, and is also provided forcomparison purposes. The return/risk ratio (39.81%) is higher than inthe previous three variations of the model that invests in only totalmarket indexes. The fifth variation expands sector investments into 80%U.S. sectors and 20% selected from approximately 170 developedmarket sectors. The return/risk ratio (54.09%) is a substantialimprovement over the U.S. sectors only portfolio (39.81%). The finalvariation constrains 80% in U.S. sectors, and 20% selected fromapproximately 170 developed market sectors and 200 emerging marketsectors. There is another large increase in the return/risk ratio to70.85%. It is worthwhile to note that the 20% invested in foreign sectorsis comprised of 18.80% emerging sectors, and 1.20% developed sectorsas determined by the optimal asset allocation model.

Table 10 contains the composition of the market and sector basedoptimized portfolios for the 1996–2000 sub period. Again, the firstmodel variation is invested 100% in the U.S. total index provided forcomparison purposes only. The second variation is restricted to invest80% in the U.S. total market index, while the model selects the totalmarket indexes of Denmark (6.41%), Finland (2.00%), France (6.59%),and Italy (5.00%) to represent the remaining 20% of the portfolio amonga choice of 17 developed equity markets. When the total market indexmodel is open to all countries, the optimal portfolio consists of


investments in U.S. (80%) and Finland (1.03%), with the remaining18.97% allocated among Chile (2%), Greece (4%), Hong Kong (0.18%),India (4%), Poland (1.72%), Portugal (4%), and Turkey (3.07%).

When sector investment is allowed, the U.S. only model selectsgeneral industrials (23%), noncyclical goods (32%), utilities (33%), andinformation technology (12%) as the optimal investment sectors. Whenall developed market sectors are allowed, eight sectors from Denmark,France, Ireland, Italy, and the U.K. are included in addition to four U.S.sectors. When all sectors (developed and emerging) are available forinvestment, the model selects 24 developed market sectors (excludingthe U.S.), four U.S. sectors, and 17 emerging sectors. The emergingsectors represent 18.8% of the portfolio. As some emerging sectors mayeither be thinly traded or dominated by only a few firms, portfoliomanagers must be cautious in making investment decisions to ensureadequate diversification.

Depending on the sample size, it appears that some sectors are moreimportant than other sectors. In the U.S. only sector selection, only foursectors are included as noted above. When the sample is expanded to alldeveloped markets, seven of the 10 sector groups are chosen. When alldeveloped and emerging sectors are available, all 10 sectors areincluded in the model. With the widest possible selection of sectorsavailable for investment, the model selects the widest variety of sectorsproducing the maximum return/risk ratio.

A similar pattern is observed in earlier periods as in the most recentperiod. In sum, the findings demonstrate that sector investments acrosscountries are superior to investing in a total market index acrosscountries regardless of the time horizon selected.(The U.S. only sectorportfolio does not outperform the total market index portfolio thatincludes developed and emerging markets in the 1991–1995 period, butdoes surpass the U.S. total market index portfolio in that period.) A U.S.investor in total market indexes or sectors will achieve greaterperformance by including foreign investments, particularly emergingmarkets. The earliest two periods (1981–1985 and 1986–1990) do notinclude emerging market investments due to data limitations. However,sector based investment between developed markets producessubstantially higher return/risk ratios than total market indexinvestment.

Once again, this evidence is consistent with the prediction of greaterdiversification benefits from investing in sectors across countries ratherthan solely in well diversified country index portfolios as posited by


Serra (2000) and Meric et al. (2001). Ex post portfolio optimizationincludes assets that stochastically dominate other assets historically.Unfortunately, ex ante knowledge of superior performing countries andsectors is unknown. The goal here is not to forecast which countries andsectors to invest in, but to simply show that sector investment is thepotentially dominant strategy of well diversified portfolios.

V. Conclusions

This paper examines the changes in international equity sector andcountry index correlations from 1981–2000, and assesses the impact onportfolio diversification benefits over time from the perspective of aU.S. investor. The correlation and panel data analyses demonstrate thattotal market index integration is rising over time. Foreign sectors arealso more highly integrated with U.S. sectors when comparing the firstsub period (1981–1985) with the last sub period (1996–2000). Paneldata tests confirm the existence of asymmetry in certain sectors, whichgenerally react more to downturns in U.S. markets than upturns.

Why are some foreign sectors more highly correlated with U.S.sectors than others? There are at least two main factors to explain this.First, the level of integration between international economies accountsfor the increase in sector and total market indexes. This is evidenced as

TABLE 10. Markowitz Mean-Variance Efficient Portfolio Optimization.Composition of Market And Sector-Based Investment Strategies inDeveloped And Emerging Markets For The Subperiod 1996–2000.

Total Market Indexes Total Market Indexes Sectors(Developed) (Developed & Emerging) (U.S. only)

80% U.S. 80% U.S. 23% Gen. Ind.6.41% Denmark 1.03% Finland 32% Non-cycl. gds2.00% Finland 2.00% Chile 33% Utilities6.59% France 4.00% Greece 12% Info. Tech.5.00% Italy 0.18% H. Kng

4.00% India1.72% Poland4.00% Portugal3.07% Turkey

(Continued)


the U.S. is more highly correlated with developed markets comparedwith emerging markets. The dramatic increase in correlations betweenU.S. and emerging markets during the 1990’s also reflects the increasein trade and investments between these entities. Second, some sectorsare impacted more from local rather than global factors. For example,information technology firms tend to trade in line with each other bothnationally and internationally according to global demand for theirproducts. Utilities, for the most part, depend more on domestic factorssuch as local consumption and government policy. Decisions on publicexpenditure, employment policies, and tax systems all continue tosegment markets to an extent.

Since the level of correlation is a significant determinant of thebenefits of international diversification, a portfolio optimization model

TABLE 10. (Continued)

Sectors(U.S. & Developed) Sectors (U.S., Developed & Emerging)

18.40% Gen. Ind. (U.S.) 0.03% ncycl. gds (Belgium) 15.25% gen. ind. (U.S.)25.6% Non-cycl. Goods (U.S.) 0.02% cycl. srv (Belgium) 34.06%ncycl. gds (U.S.)26.40% Utilities (U.S.) 0.05% ncycl. srv. (Canada) 20.59% utilities (U.S.)9.60% Info. Tech. (U.S.) 0.07% financials (Canada) 10.10% info. tech. (U.S.)5.00% Gen. Ind. (Denmark) 0.06% basic ind. (Denmark) 0.34% basic (Greece)1.38% Non-cycl. Gds (Denmark) 0.08% cycl. gds (Denmark) 3.95% cycl. srv (Greece)0.21% Financials (Denmark) 0.07% ncycl. srv. (Dnmrk) 1.19%info.tech. (Greece)0.14% Non-cycl. Gds (France) 0.06% basic ind. (France) 0.38%financials (Greece)0.32% Non-cycl. Srvcs (France) 0.03% gen. ind. (France) 0.60% resource (H. K.)4.14% Cycl.-Srvcs (Ireland) 0.10% info. tech (France 2.00% utilities (H. Kong)3.81% Non-cycl. Srvcs (Italy) 0.01% financials (France) 0.60% info. tech. (H. K.)5.00% Info. Tech. (U.K.) 0.07% basic ind.(Germany) 0.22% basic ind. (India)

0.09% gen. ind.(Germany) 0.45% ncycl. gds (India)0.05% cycl. srv. (Germany) 0.20% cycl. srv (India)0.05% basic ind. (Ireland) 0.80% utilities (India)0.05% cycl. gds. (Ireland) 2.72% info. Tech. (India)0.01% cycl. gds. (Italy) 0.10% financials (India)0.10%resource(Nethlnd) 0.60%resource(N.Z.)0.06% cycl. gds. (Nethlnd) 0.62% gen. ind. (N. Z.)0.01% financials (Norway) 0.43% cycl. gds. (N. Z.)0.02% basic ind. (Norway) 3.60%cycl. srv. (Taiwan)0.02% basic ind. (Swtzerlnd)0.05% financials (Swtzerlnd)0.05% resource (UK)


is utilized to demonstrate the value of fundamental sector analysis inforeign investment to U.S. investors. The model assumes the position ofa typical U.S. pension fund that invests 80% in the U.S. and 20%internationally. Several variations of the model are tested thatspecifically include or exclude total market indexes, sector onlyinvestments, investments in developed markets, and investments inemerging markets.

The results clearly indicate on an ex post basis the superiority ofasset allocation strategies that utilize sector based investing acrosscountries compared with total market index investments. Also,portfolios that include investment in emerging markets provide superiorreturn/risk ratios than portfolios that only invest in developed markets.

Although correlations between U.S. and most other markets andsectors have increased dramatically over the past 20 years, carefulsector or total market index investment may provide significantinternational diversification benefits to a U.S. investor’s portfolio. Asthis procedure is performed on an ex post basis, it is not appropriate touse these portfolio weights in future investments. Dahlquist and Harvey(2001) provide a strategy for a forward-based portfolio model.

References

Aggarwal, R. and Leal R. 1997. Linkages and volatility in emerging capitalmarkets. In M. Papaioannou and G. Tsetsekos (eds). Emerging MarketPortfolios: Diversification and Hedging Strategies, Chicago: Irwin137–149.

Baca, S., Garbe B. and Weiss R. 2000. The rise of sector effects in major equitymarkets. Financial Analysts Journal 56: 34–40.

Basu, P. and Morey, M.R. 2000. Liberalization and stock prices in emergingmarkets. Emerging Markets Quarterly (Fall): 1–17.

Beckers, S.; Connor G. and Curds, R. 1996. National versus global influenceson equity returns. Financial Analysts Journal 52: 31–39.

Bekaert, G. 1995. Market integration and investment barriers in emergingequity markets. World Bank Economic Review 9: 75–107.

Bekaert, G. and Harvey, C. 2000. Foreign speculators and emerging equitymarkets. Journal of Finance 55: 565–613.

Bekaert, G. and Harvey, C. 1997. Emerging equity market volatility. Journal ofFinancial Economics 43: 29–77.

Dahlquist, M. and Harvey, C. 2001. Global tactical asset allocation. EmergingMarket Quarterly 5: 6–14.

Divecha, A. B.; Drach, J. and Stefek, D. 1992. Emerging markets: Aquantitative perspective. Journal of Portfolio Management (Fall): 41–50.


Erb, C.; Harvey, C. and Viskanta, T. 1995. Do world markets still serve as ahedge? Journal of Investing 4: 1–17.

Erb, C.; Harvey, C. and Viskanta, T. 1994. Forecasting internationalcorrelation. Financial Analysts Journal 50: 32–45.

Goetzmann, W. N.; Li, L. and Rouwenhorst, G. 2005. Long-term global marketcorrelations. Journal of Business,78:1-38.

Greene, W. 1990. Econometric Analysis, New York: MacMillan Publishing.Harvey, C. 1995. Predictable risk and returns in emerging markets. Review of

Financial Studies 8: 773–816. Henry, P. 2000. Stock market liberalization, economic reform, and emerging

market equity prices. Journal of Finance 55: 529–564. Heston, S. and Rouwenhorst, G. 1994. Does industrial structure explain the

benefits of international diversification? Journal of Financial Economics 36: 3–27.

Kawakatsu, H. and Morey, M. R. 1999. An empirical examination of financialliberalization and the efficiency of emerging market stock prices. Journalof Financial Research 22: 385–411.

Levy H. and Sarnat, M. 1970. International diversification of investmentportfolios. American Economic Review 60: 668–675.

Longin, F. and Solnik, B. 1995. Is the correlation in international equity returnsconstant: 1960–1990? Journal of International Money and Finance 14:3–26.

Meric, I. and Meric, G. 1998. Correlation between the world’s stock marketsbefore and after the 1987 crash. Journal of Investing 7: 67–70.

Meric, G.; Leal, R.; Meric, I. and Ratner, M. 2001. Co-movements of U.S. andLatin American equity markets before and after the 1987 crash.International Review of Financial Analysis 10: 219–235.

Miller, C. H. 2002. Morgan Stanley Capital International. Newswatch, p. 4.Roll, R. 1992. Industrial structure and the comparative behavior of international

stock market indexes. Journal of Finance 47: 3–42.Rouwenhorst, G. 1999. European equity markets and the EMU. Financial

Analysts Journal 55: 57–64Serra, A. P. 2000. Country and industry factors in returns: Evidence from

emerging markets’ stocks. Emerging Markets Review 1:127–151. Solnik, B. 1974. Why not diversify internationally? Financial Analysts Journal

30: 48–54.Solnik, B., Boucrelle, C. and Fur, Y. 1996. International market correlation and

volatility. Financial Analysts Journal 52: 17–34.

Documents

Sector Integration and the Benefits of Global Diversification